In our paper we propose some techniques to detect Unmanned Aerial Vehicles (UAVs), aka drones, in a real life scenario, such as airports. Using the dataset of Artem Rozantsev (a collection of drones and no- drones sample patches) we try to investigate different approaches to overcome such a challenging problem to identify flying objects in different background scenarios and recognize them to be drones or not.
To train the network as explained in the paper you need to download the dataset uav200 from here and put "data" into data folder (or your own samples). Then launch main.py file.
To use the network, divide the video you want to test into jpg frames and load them into video_test folder. Then launch sliding_window.py (change weights accordingly for predictions if needed). You can download our weights from here
